[Q] What is the significance of the "Out of IOMMU Space" error?

[Q] What is the significance of the "Out of IOMMU Space" error?

[Maurice Volaski]

Running 2.6.12 on AMD Opteron on an Arima HDAMA board with a 3ware 
8506 card installed and got the following in the log:

Jun 25 15:56:34 [kernel] [53161.700276] PCI-DMA: Out of IOMMU space 
for 65536 bytes at device 0000:03:01.0
Jun 25 15:56:34 [kernel] [53161.700279] 3w-xxxx: 
tw_map_scsi_sg_data(): pci_map_sg() failed.
Jun 25 15:56:34 [kernel] [53161.700332] SCSI error : <0 0 0 0> return 
code = 0x70000
Jun 25 15:56:34 [kernel] [53161.700334] end_request: I/O error, dev 
sda, sector 51296795


This error doesn't seem to be mentioned much in this space. Should it 
be taken at face value? Is it something simple like IOMMU should be 
set to a higher value in the HDAMA bios? A kernel bug? Or could it 
actually be bad physical memory or even a bad card?

Does it mean the data integrity on the affected hard drive is compromised?
-- 

Maurice Volaski, mvolaski@aecom.yu.edu
Computing Support, Rose F. Kennedy Center
Albert Einstein College of Medicine of Yeshiva University

Re: [Q] What is the significance of the "Out of IOMMU Space" error?

[David S. Miller]

From: Maurice Volaski <mvolaski@aecom.yu.edu>
Date: Sun, 26 Jun 2005 00:37:08 -0400

> This error doesn't seem to be mentioned much in this space. Should it 
> be taken at face value? Is it something simple like IOMMU should be 
> set to a higher value in the HDAMA bios? A kernel bug? Or could it 
> actually be bad physical memory or even a bad card?

It could be a DMA mapping leak in some kernel driver.
Or it could be that there is more DMA in flight than
the IOMMU space available can hold.

Re: [Q] What is the significance of the "Out of IOMMU Space" error?

[Andi Kleen]

Maurice Volaski <mvolaski@aecom.yu.edu> writes:
> 
> This error doesn't seem to be mentioned much in this space. Should it
> be taken at face value? Is it something simple like IOMMU should be
> set to a higher value in the HDAMA bios? A kernel bug? 

It could be the second, but normally it is the first. The defaults
supplied by the BIOS tend to be very tight and the 3ware controller
eats a lot of IOMMU space because it has deep queues.  I would just
increase the IOMMU space in the BIOS.

-Andi